外文翻譯--基于模煳測量和模煳積分融合技術(shù)的機(jī)械故障診斷 (英文）

1、Machinery fault diagnosis based on fuzzy measure and fuzzy integral data fusion techniquesXiaofeng Liu ?, Lin Ma, Joseph MathewCRC for Integrated Engineering Asset Management, School of Engineering Systems, Queensland Un

2、iversity of Technology,GPO Box 2434, Brisbane QLD 4001, Australiaa r t i c l e i n f oArticle history:Received 23 July 2007Received in revised form19 July 2008Accepted 27 July 2008 Available online 3 August 2008Keywords:

3、Fuzzy measuresFuzzy integralsFuzzy c-meansData fusionFault diagnosisa b s t r a c tFuzzy measure and fuzzy integral theory are an outgrowth of classical measure theory.Fuzzy measure and fuzzy integral theory take into ac

4、count the importance of criteriaand interactions among them, and have excellent potential for applications such asclassification. This paper presents a novel data fusion approach for machinery faultdiagnosis using fuzzy

5、measures and fuzzy integrals. The approach consists of a feature-level data fusion model and a decision-level data fusion model. The fuzzy c-meansanalysis method was employed to identify the relations between a feature s

6、et and afault prototype to establish mappings between features and given faults. Rollingelement bearing and electrical motor experiments were conducted to validate themodels. Different features were obtained from recorde

7、d signals and then fused at bothfeature and decision levels using fuzzy measure and fuzzy integral data fusiontechniques to produce diagnostic results. The results showed that the proposedapproach performs very well for

8、bearing and motor fault diagnosis.fax: +612 49201401.E-mail addresses: xf.liu@qut.edu.au, xiaofeng.liu@downeredirail.com.au (X. Liu).Mechanical Systems and Signal Processing 23 (2009) 690–700The function value f(xi) here

9、 can be interpreted as an evaluation of the confidence level of information source xi to a specific object, the fuzzy measure gl or m can be interpreted as the importance or contribution of information source xi to the f

10、inal evaluation or decision-making. This research will employ average membership value or fault recognition rates of FCM analysis for different faults using training data as fuzzy densities to construct gl fuzzy measures

11、. The recognition rate is defined as the ratio of the number of a specific fault recognized to the total number of faultsR ¼ nr n (4)where R represents the recognition rate, nr denotes the number of faults recognize

12、d correctly, n is the total fault number. The fault recognition rates or average membership values reflect the importance or contribution of feature sets or classifiers to a specific fault.2.2. Feature-level fuzzy integr

13、al diagnosis modelFig. 1 illustrates the architecture of the feature-level fuzzy integral fusion model for machinery fault diagnosis. The model consists of three major modules: the partial matching module, interaction an

14、d importance inference module and the global matching module. The first step of the feature-level fuzzy integral fusion for diagnosis is to obtain a degree of partial matching which is needed by the fuzzy integrals as f-

15、function. The partial match with respect to a feature is the determination of a partial matching degree between a feature value and a fault prototype to establish the relation between a feature value and a given fault. D

16、ifferent methods can be used to build these partial matching relations, e.g., the probability density function method. The work presented in this paper employed the FCM analysis method to identify the partial matching re

17、lations. The matching degree was represented by fuzzy membership degrees. The f-function represents the direct evidence that a fault prototype belongs to a category from the standpoint of individual information source [2

18、3]. The second step of implementation of this model is the identification of fuzzy measures. This work employed gl fuzzy measures. To obtain these gl fuzzy measures, the average membership degrees of different features f

19、or different fault prototypes were used as fuzzy densities. The fuzzy density reflects the overall confidence level of a feature for the recognition of a given fault prototype. As the fuzzy densities have been derived fr

20、om FCM membership degrees, gl can be determined bygl ¼ 1 lY ni¼1 ð1 þ lgiÞ ? 1“ #(5)where gi is the fuzzy density. l can be obtained by solving the following equation:l þ 1 ¼ Y ni¼

21、1 ð1 þ lgiÞ (6)ARTICLE IN PRESSFuzzy integral fusion f Final ResultsPartial matching 1 Partial matching 2 Partial matching n Feature 1 Feature n Partial matching Global matching Fuzzy densities Feature